Skip to content

Oil-Film Thickness

The oil stage turns the raw oil-film traverse (oil_thickness/data) into a fixed-length, dropout-free profile on the integer millimeter grid, one row per position 0–199 mm. It runs as part of rddac preprocess; every parameter below can be overridden via --config.

Property Raw Processed
Shape variable, ~(420, 2) at BF 100 kN (two readings per position), ~(210, 2) otherwise fixed (200, 2) float32
Columns sensor_position, oil_value unchanged
Position range 0 to ~210 mm 0 to 199 mm (integer grid)
Units mm, g/m² unchanged
Invalid values NaN dropouts none (removed + interpolated)

Why this stage exists

About two thirds of the experiments contain sensor dropouts: isolated NaN readings, typically one to six per trace, almost always at single positions. A naive robust filter is blinded by them: a NaN inside a Hampel window makes the window median NaN, silently disabling outlier detection within ±5 mm of every dropout, precisely where spikes tend to appear. The published files carry the data exactly as recorded, so this cleaning belongs to preprocessing.

Processing steps

  1. Truncation: positions at/after 200 mm are edge artifacts beyond the part and are dropped.
  2. Dropout removal: NaN readings are removed before filtering so they cannot poison the Hampel windows.

  3. Hampel filter: each reading is compared against its position neighborhood (±5 mm). A reading is an outlier when

    \[ |x_i - \tilde{x}_i| > k \cdot \sigma_i, \qquad \sigma_i = 1.4826 \cdot \operatorname{MAD}_i \]

    where \(\tilde{x}_i\) is the window median and \(k = 3\). The sensor logs two decimals, so flat windows can collapse to \(\operatorname{MAD} = 0\); the scale then falls back to the mean absolute deviation, floored at one logging quantization step \(q = 0.01\ \text{g/m}^2\):

    \[ \sigma_i = \max\!\left(\sqrt{\tfrac{\pi}{2}}\; \overline{|x - \tilde{x}_i|},\; q\right) \]

    so isolated spikes are still caught while ±1 LSB wiggle is never flagged.

  4. Duplicate averaging: BF 100 kN experiments measure each position twice; remaining readings are averaged per integer position.

  5. Grid fill: the integer 0–199 mm grid is completed by linear interpolation, with nearest-value fill at the edges.

Oil-film processing of one experiment: raw readings with the 200 mm truncation line (left), Hampel-flagged outliers (middle), the final fixed-grid profile (right).

Parameters

TOML key ([oil]) Default Meaning
max_sensor_position 200 positions at/after this (mm) are edge artifacts
output_length 200 fixed output row count
hampel_window 5 half-window in mm
hampel_k 3.0 threshold in robust sigmas
value_quantization 0.01 logging LSB (g/m²), floors the flat-window scale

Every value used is stamped into the oil_thickness group attributes together with the cleaning counts (n_nan_removed, n_hampel_outliers, n_positions_interpolated), so a processed file stays self-describing.

Further reading